基于自联想神经网络的数据滤波与应用  

Data filtering method and application based on auto-associative neural network

作　　者：才轶[1] 徐圆[1] 朱群雄[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029

出　　处：《计算机与应用化学》2009年第5期673-676,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金项目(60774079);国家863计划(2006AA04Z184)

摘　　要：石油化工过程系统及其现场数据复杂,基于数据驱动的任何研究、设计、运行工作首先都需要进行数据滤波。本文研究了用自联想神经网络对化工过程数据进行滤波的方法。自联想神经网络通过使输入节点的信息压缩在隐层节点上,从网络输入的高维参数空间中提取反映系统结构的最具代表性的低维子空间,同时有效地滤去了测量数据中的噪声和测量误差,再通过输出层实现数据的解压缩,将前面压缩的信息还原到各个参数值,从而实现各测量数据的重构。通过对测试函数的应用和误差比较验证了该方法可以达到比较理想的滤波效果,并采用该方法对某企业精对苯二甲酸(PTA)工业数据进行滤波后BP建模,该模型的预测效果要大大好于没有进行数据滤波建立的模型,从而进一步说明了用自联想神经网络对工业数据滤波不但是可行且有效的,同时也提高了模型预测的准确性。Because of the chemical process system and its data are complicated, all study, design and running based on data-driven need data filtering. A method of industrial data filtering by using auto-associative neural network （AANN） is studied in this paper. The AANN extracts the most representative low-dimensional sub-space characteristic which reflects the system structure from high-dimensional parameter space of network input by compressing input information nodes in the hidden layers, while noise and measuring errors are filtered effectively, and data decompression through the output layer. So the reconstruction of information is realized. Firstly, the method has been validated through the application of the test functions and error comparisons. Moreover, it is used to the data filtering and BP modeling of the industrial PTA production system. Forecasting precision of the model is much better than that is no data filtering. The experimental results show that the method of chemical data filtering based on AANN is feasible and effective, and the accuracy of model prediction can be improved.

关 键 词：自联想神经网络 BP算法 噪声 数据滤波 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]